Abstract: A rapid quantitative model of 4 kinds of flour components was established by using near-infrared spectroscopy combined with different quantitative analysis methods. The content of water, fat, carbohydrate and protein in 68 flour samples was determined by GB method, and its near-infrared diffuse reflectance spectra were collected. After selecting 58 correction sets and 10 verification set samples, the quantitative model effect of partial least squares method (partial least squares, PLS) based on full spectrum was compared with 17 kinds of spectral pretreatments after eliminating abnormal samples by Mahalanobis distance method, and the optimal quantitative model was obtained by screening the characteristic spectra by using the backward interval partial least squares method (Backward interval PLS, BiPLS) on the basis of the best preprocessing method. The results showed that the correlation coefficient Rcv of the model correction set was greater than 0.9650, the mean square root error of the internal cross verification was less than 0.328, the correlation coefficient of the verification set was greater than 0.9926, and the prediction mean square root error is less than 0.383.Therefore, the model would have good accuracy and stability, and it could be applied to the multi-index rapid detection of flour.